EP0415000B1 - Method and apparatus for spelling error detection and correction - Google Patents

Method and apparatus for spelling error detection and correction Download PDF

Info

Publication number
EP0415000B1
EP0415000B1 EP19900108508 EP90108508A EP0415000B1 EP 0415000 B1 EP0415000 B1 EP 0415000B1 EP 19900108508 EP19900108508 EP 19900108508 EP 90108508 A EP90108508 A EP 90108508A EP 0415000 B1 EP0415000 B1 EP 0415000B1
Authority
EP
European Patent Office
Prior art keywords
word
information signal
information
words
means
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Lifetime
Application number
EP19900108508
Other languages
German (de)
French (fr)
Other versions
EP0415000A2 (en
EP0415000A3 (en
Inventor
Frederick J. Damerau
Eric K. Mays
Robert L. Mercer
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
International Business Machines Corp
Original Assignee
International Business Machines Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority to US07/401,584 priority Critical patent/US5258909A/en
Priority to US401584 priority
Application filed by International Business Machines Corp filed Critical International Business Machines Corp
Publication of EP0415000A2 publication Critical patent/EP0415000A2/en
Publication of EP0415000A3 publication Critical patent/EP0415000A3/en
Application granted granted Critical
Publication of EP0415000B1 publication Critical patent/EP0415000B1/en
Anticipated expiration legal-status Critical
Application status is Expired - Lifetime legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/20Handling natural language data
    • G06F17/27Automatic analysis, e.g. parsing
    • G06F17/273Orthographic correction, e.g. spelling checkers, vowelisation

Description

  • The invention relates to a method and an apparatus for detecting an error in an information signal.
  • In text processing apparatus, such as dedicated word processors or word processing programs which are run on general purpose digital computers, it is desirable to provide automatic detection and correction of spelling errors. Most spelling error detection apparatus and programs check each word in a text against the entries in a spelling dictionary. Words in the text which are not found in the spelling dictionary are assumed to be misspelled. The misspelled words are identified to the text processing operator by, for example, highlighting the word on a display device. Sometimes candidate words having spellings similar to the misspelled word are also displayed to the operator as proposed corrections.
  • Other text processing systems are known from AFIPS Conference Proceedings 1982 National Computer Conference, 7 June 1982, Houston, Texas, US, pp. 501 - 508, S.N. Srihari et. al.: "Integration of botton-up and top-down contextual knowledge in text error correction" and IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. PAMI8, No. 6, November 1986, New York, US, pp. 742 - 749, A.M. Derouault & B. Merialdo: "Natural language modeling for phenome-to-text transcription".
  • The known apparatus and methods for detecting and correcting spelling errors have several deficiencies. Most importantly, the known apparatus and methods cannot detect a "wrong word" erroneous spelling (where the erroneous spelling is itself a word in the spelling dictionary but is not the word that was intended).
  • Moreover, even where the erroneous spelling does not appear in the spelling dictionary, the prior apparatus and methods provide no means or only limited means for ranking alternative candidates for the correct spelling.
  • Summary of the Invention
  • It is an object of the invention to provide a method and apparatus for detecting and correcting an error in an information signal, where the information signal represents the wrong information. When the information signal represents a word, the invention provides a method and apparatus for detecting and correcting spelling errors, where erroneously spelled words are correct entries in the spelling dictionary, but are not the intended words.
  • It is another object of the invention to provide a method and apparatus for estimating the probability of occurrence of a word whose spelling is being checked, and to estimate the probabilities of one or more alternative words as candidates for replacing the word being checked.
  • In a spelling error detection and correction method according to the present invention, an input string of words Wi is provided. The spelling of a first word W1 in the input string is changed to form a second word W2 different from the first word, to form a candidate string of words Wc. The probability P(Wi) of occurrence of the input string of words and the probability P(Wc) of occurrence of the candidate string of words are estimated. The probability P(Wi|Wc) of misrepresenting the candidate string of words Wc as the input string of words Wi is also estimated. Thereafter, P(Wi) is compared with the product P(Wc)P(Wi|Wc). A first output is produced if P(Wi) is greater than P(Wc)P(Wi|Wc), otherwise a second output is produced.
  • In one aspect of the invention, the first output comprises the input string of words. The second output, comprises the candidate string of words. Alternatively, the second output may be an error indication.
  • The probability P(Wi|Wc) of misrepresenting the candidate string of words as the input string of words may be estimated as the probability P(W1|W2) of misspelling the second word W2 as the first word W1.
  • In the spelling error detection and correction method and apparatus according to the invention, each word in the input string and each word in the candidate string is a member of a set of correctly spelled words.
  • Preferably, the method and apparatus according to the invention further comprise the step of estimating the probability P(Wi|Wi) of correctly spelling all of the words in the input string of words W1 In this case, the product P(Wi)P(Wi|Wi) is compared with the product P(Wc)P(Wi|Wc). The first output is produced if P(Wi)P(Wi|Wi) is greater than P(Wc)P(Wi|Wc), otherwise the second output is produced.
  • The probability P(Wi|Wi) of correctly spelling all of the words in the input string may be estimated as the probability P(W1|W1) of correctly spelling the first word W1.
  • According to an embodiment of the invention, the spelling of the first word W1 may be changed to form the second word W2 by adding, deleting, transposing, or replacing one or more letters in the first word to form a tentative word. The tentative word is compared to each word in the set of words. The tentative word is used as the second word W2 if the tentative word matches a word in the set of correctly spelled words.
  • Alternatively, the spelling of the first word may be changed to form a second word by identifying a confusion group of M different words in the set of correctly spelled words. Each word in the confusion group may, for example, have a spelling which differs from the first word by no more than two letters. Alternatively, each word in the confusion group may be one which is confusable with the first word. At least one word in the confusion group is selected as the second word W2.
  • Satisfactory results have been obtained in the method and apparatus according to the invention by estimating the probability of correctly spelling a word as 0.999. The probability of misspelling a word may be estimated to be 0.001 M
    Figure imgb0001
    .
  • The spelling error detection and correction method and apparatus according to the present invention are advantagecus because by comparing the probability of occurrence of the word being checked and the probabilities of occurrence of one or more spelling correction candidates, it is possible to detect and correct errors which are correct spellings of the wrong word.
  • Brief Description of the Drawing
  • FIG. 1 is a block diagram of an embodiment of the spelling error detection and correction method according to the present invention.
  • FIG. 2 is a block diagram of an example of the spelling error detection and correction method of FIG. 1.
  • FIG. 3 is a block diagram of another example of the spelling error detection and correction method of FIG. 1.
  • FIG. 4 is a block diagram of an embodiment of a routine for changing the spelling of a first word to form a second word in the spelling error detection and correction method according to the present invention.
  • FIG. 5 is a block diagram of an alternative embodiment of a method of changing the spelling of a first word to form a second word.
  • FIG. 6 is a block diagram of a preferred modification of the spelling error detection and correction method shown in FIG. 1.
  • FIG. 7 is a block diagram of an embodiment of an apparatus for detecting and correcting an error in an information signal.
  • Description of the Preferred Embodiments
  • The invention is a method of detecting and correcting an error in an information signal. In the case where each information signal represents a word which is a member of a set of correctly spelled words, the invention provides a method of spelling error detection and correction.
  • Referring to FIG. 1, the spelling error detection and correction method starts with the step of providing an input string of words Wi. Each word in the input string has a spelling.
  • Next, the spelling of a first word W1 in the input string is changed to form a second word W2 different from the first word, to form a candidate string of words Wc.
  • In FIG. 1, the input string and the candidate string each comprise three words. According to the invention, the input and candidate strings may be any length greater than or equal to two. Each string may be, for example, a sentence or a phrase.
  • Next, the probabilities P(Wi) of occurrence of the input string of words and P(Wc) of occurrence of the candidate string of words are estimated. These probabilities may be estimated empirically by examining large bodies of text, as discussed in more detail, below.
  • Also estimated is the probability P(Wi|Wc) of misrepresenting the candidate string of words Wc as the input string of words Wi. The probability P(WiWc) may be chosen empirically by selecting different values until satisfactory results are obtained, as discussed in the Examples below.
  • After the required probabilities are estimated, the probability P(Wi) is compared with the product of the probabilities P(Wc)P(Wi|Wc). If P(Wi) is greater than or equal to the product P(Wc)P(Wi|Wc), then a first output is produced. Otherwise, a second output is produced.
  • As shown in FIG. 1, the first output may be the input string W1WMWN. The second output may be the candidate string W2WMWN.
  • Alternatively, the second output may be an error indication.
  • Two examples of the spelling error detection and correction method according to the present invention are shown in FIGS. 2 and 3. Referring to FIG. 2, the input string is a string of three words: "the horse ran". Each word in the input string of words is a member of a set of correctly spelled words. The first word W1 is "horse".
  • Next, the spelling of the first word "horse" is changed to form the second word W2, "house". The candidate string of words Wc is then "the house ran". The second word "house" is also a member of the set of correctly spelled words.
  • Continuing, the probability P(Wi) of occurrence of the input string of words "the horse ran" is estimated to be 5x10-5. The probability P(Wc) of occurrence of the candidate string of words "the house ran" is estimated to be 1x10-8. While these probabilities are purely hypothetical for the purpose of illustrating the operation of the present invention, the hypothetical numbers illustrate that the probability of occurrence of "the horse ran" is much greater than the probability of occurrence of "the house ran".
  • Proceeding with the method, the probability P(Wi|Wc) of misrepresenting the candidate string of words as the input string of words is estimated to be equal to the probability P(W1|W2) of misspelling the second word W2 as the first word W1 From experiment it has been determined that an estimate of 0.001 produces satisfactory results.
  • Finally, the value of P(Wi) is compared to the product P(Wc)P(Wi|Wc). Since the former (5x10-5) is greater than the latter (1x10-11), the input string of words is determined to be correct, and the candidate string of words is rejected. Accordingly, the output is "the horse ran".
  • FIG. 3 illustrates the operation of the spelling error detection and correction method where the input string is "the house ran". Now the first word W1 is "house," and the second word W2 is "horse". By using the same probabilities estimated in FIG. 2, the probability of the input string (1x10-8) is now less than the product of the probability of the candidate string multiplied by the probability of misrepresenting the candidate string as the input string (5x10-8). Therefore, the input string is now rejected, and the candidate string is determined to be correct. The output is set to "the horse ran".
  • The spelling error detection and correction method according to the present invention is based on the following theory. For each candidate string of words (for example, for each candidate sentence) Wc, the probability that the candidate sentence was actually intended given that the original sentence (input string of words) Wi was typed is given by P(W c |W i ) = P(W c )P(W i |W c ) P(W i ) .
    Figure imgb0002
    In this equation, P(Wi|Wc) is the probability of misrepresenting the candidate string of words Wc as the input string of words Wi.
  • The probability P(Wi|Wi) that the original sentence was actually intended given that the original sentence was typed (that is, the probability of correctly spelling all of the words in the original sentence Wi) is compared to P(Wc|Wi). For simplicity, both sides of the comparison are multiplied by P(Wi) so that the product P(Wi)P(Wi|Wi) is compared with the product P(Wc)P(Wi|Wc) The sentence with the higher probability is selected as the sentence which was actually intended.
  • In order to further simplify the comparison, it may be assumed that the probability P(Wi|Wi) that the original sentence was actually intended given that the original sentence was typed is equal to 1.
  • The probabilities P(Wi) of occurrence of the input string of words and P(Wc) of occurrence of the candidate string of words may be approximated by the product of n-gram probabilities for all n-grams in each string. That is, the probability of a string of words may be approximated by the product of the conditional probabilities of each word in the string, given the occurrence of the n-1 words (or absence of words) preceding each word. For example, if n = 3, each trigram probability may represent the probability of occurrence of the third word in the trigram, given the occurrence of the first two words in the trigram.
  • The conditional probabilities may be determined empirically by examining large bodies of text. For example, the conditional probability
    Figure imgb0003
    (Wz|WxWy) of word Wz given the occurrence of the string WxWy may be estimated from the equation
    Figure imgb0004
    where f 1 (W z |W x W y ) = n xyz n xy
    Figure imgb0005
    f 2 (W z |W y ) = n yz n y
    Figure imgb0006
    f 3 (W z ) = n z n
    Figure imgb0007
    f 4 = 1 n
    Figure imgb0008
    and λ 1 + λ 2 + λ 3 + λ 4 = 1
    Figure imgb0009
  • In equations (3)-(6), the count nxyz is the number of occurrences of the trigram WxWyWx in a large body of training text. The count nxy is the number of occurrences of the bigram WxWy in the training text. Similarly, nyz is the number of occurrences of the bigram WyWz in the training text, ny is the number of occurrences of word Wy, nz is the number of occurrences of word Wz, and n is the total number of words in the training text. The values of the coefficients λ123, and λ4 in equations (2) and (7) may be estimated by the deleted interpolation method described in an article by Lalit R. Bahl et al entitled "A Maximum Likelihood Approach to Continuous Speech Recognition" (IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. PAMI-5, No. 2, March 1983, pages 179-190).
  • In the comparison of P(Wi)P(Wi|Wi) with P(Wc)P(Wi|Wc) the probability P(Wi|Wc) may be approximated by the product of the probabilities of misrepresenting each word in the candidate sentence as the corresponding word in the original typed sentence. Where the original typed sentence and the candidate sentence differ by only one word (W1 in the original sentence and W2 in the candidate sentence), the probability P(Wi|Wc) can be estimated to be equal to the probability P(W1|W2) of misspelling the second word as the first word.
  • The probability of misspelling any given word should be estimated to have a low value, for example 0.001. This value has been determined by experiment to yield satisfactory results. By increasing the probability of misspelling, the invention will find more misspellings; by decreasing the probability of misspelling, the invention will find fewer misspellings. When the word W1 in the original typed sentence has M misspellings which result in correct dictionary entries, the probability of each misspelling becomes 0.001 M
    Figure imgb0010
    in this example.
  • If the probability P(Wi|Wi) of correctly spelling all of the words in the original typed sentence is not estimated as 1, it may be approximated by the product of the probabilities of correctly spelling each word in the original typed sentence. Where the original typed sentence and the candidate sentence differ by only one word, the probability P(Wi|Wi) may be estimated as the probability P(W1|W1) of correctly spelling the first word W1.
  • FIG. 4 shows a subroutine which may be used to change the spelling of the first word W1 to the second word W2. First, one or more letters in the first word W1 are changed to form a tentative word WT. The changes may be made by, for example, adding a letter to the first word, deleting a letter from a first word, transposing two letters in the first word, or replacing a letter in a first word.
  • The tentative word WT is then compared to each word in a set of words (a spelling dictionary) L. If the tentative word WT matches a word in the spelling dictionary L, then the second word W2 is set equal to the tentative word.
  • FIG. 5 shows an alternative subroutine for changing the spelling of a word. In this routine, each word in the spelling dictionary is provided with an associated confusion group of words Lc containing M different words. For example, each word in the confusion group may have a spelling which differs from the spelling of the first word W1 by no more than two letters. Alternatively, each word in a confusion group may be a word which sounds like and is therefore confusable with the first word (for example, "to", "two", and "too", or "principle" and "principal"). For each candidate sentence, one word is selected from the confusion group Lc as the second word W2.
  • FIG. 6 shows a modification of the spelling error detection and correction method of FIG. 1. The steps shows in FIG. 6 are intended to replace the steps in block 6 of FIG. 1.
  • According to the modification, the method further includes the step of estimating the probability P(Wi|Wi) of correctly spelling all of the words in the input string of words Wi. The product of P(Wi)P(Wi|Wi) is compared with the product P(Wc)P(Wi|Wc). If the former is greater than or equal to the latter, a first output (for example, the input string) is produced. If the former is less than the latter, then a second output (for example, the candidate string) is produced.
  • An apparatus for detecting and correcting an error in an information signal, for example where each information signal represents a word having a spelling, is preferably in the form of a programmed general purpose digital computer. FIG. 7 shows an example of the organization of such an apparatus.
  • As shown in FIG. 7, a word processor 10 provides an input string of information signals Wi. Each information signal represents information, such as a word. The word processor 10 is preferably a program running on a central processor 12 which is also executing the other functions of the apparatus. However, word processor 10 may alternatively be running on its own central processor.
  • Under the direction of the program instructions in program instructions store 14, the central processor 12 changes a first information signal W1 in the input string Wi to form a second information signal W2 representing information which is different from the information represented by the first information signal. This change forms a candidate string of information signals Wc. Under the direction of the program instructions, central processor 12 compares the second information signal W2 with the entries in the spelling dictionary store 16 to be sure that the second information signal is an entry in the spelling dictionary.
  • Having produced the input and candidate strings, central processor 12 is instructed to retrieve estimates of the probabilities of occurrence of the input and candidate strings from the word string probability store 18. The probability P(Wi|Wc) of misrepresenting the information represented by the candidate string of information signals as the input string of information signals is retrieved from store 20. Finally, central processor 12 compares P(Wi) with the product P(Wc)P(Wi|Wc). A first output signal is sent to, for example, a display 22 if the former is greater than or equal to the latter. Otherwise, a second output signal is sent to the display 22.
  • The spelling error detection and correction method and apparatus according to the present invention were tested on 3,044 sentences which were systematically misspelled from 48 sentences. The 48 sentences were chosen from the Associated Press News Wire and from the Proceedings of the Canadian Parliament. Trigram conditional probabilities were obtained from a large corpus of text consisting primarily of office correspondence. Using a probability P(Wi|Wi) of 0.999, the method selected the changed sentence 78% of the time. Of those sentences that were changed, they were changed correctly 97% of the time.
  • Several examples selected from the above-described tests are described below.
  • EXAMPLE I
  • In this example, the input word string (the original typed sentence) is "I submit that is what is happening in this case." The word W1 whose spelling is being checked is "I". The word "I" has only the following simple misspelling: "a". Therefore, the second word W2 is "a", and the candidate word string Wc (the candidate sentence) is "a submit that is what is happening in this case."
  • Table 1 shows the input and candidate sentences, the trigrams which make up each sentence, and the natural logarithms of the conditional probabilities for each trigram. The experiment was performed with four different values of the probability Pt of correctly spelling each word: Pt = 0.9999, Pt = 0.999, Pt = 0.99, or Pt = 0.9.
  • Since the logarithms (base e) of the probabilities are estimated in Table 1, the logarithms are added to produce estimates of the product of the probabilities.
  • Table 2 shows the totals obtained from Table 1. For all values of Pt , the original sentence Wi is selected over the alternative candidate sentence Wc.
    Figure imgb0011
    Figure imgb0012
  • EXAMPLE II
  • In this example, the input word string Wi is: "I submit that is what is happening in this case". The first word W1 whose spelling is being checked is "submit". The word "submit" has two simple misspellings: "summit" or "submits". In this example, the second word W2 is selected to be "summit". Therefore, the candidate word string Wc (the candidate sentence) is "I summit that is what is happening in this case."
  • Table 3 shows the logarithms of the probabilities, and Table 4 provides the totals for Table 3. Again, for each value of Pt, the original sentence is selected over the candidate.
    Figure imgb0013
    Figure imgb0014
  • EXAMPLE III
  • In this example, the input word string Wi (the original typed sentence) is now "a submit that is what is happening in this case." The first word W1 whose spelling is being checked is "a". The word "a" has the following ten simple misspellings: "I", "at", "as", "an", "am", "ad", "ab", "pa", "or", "ha".
  • A second word W2 is selected to be "I". Therefore, the candidate string is "I submit that is what is happening in this case."
  • The logarithms of the individual probabilities are shown in Table 5. Note that the probability P(W1|W2) is equal to P t M
    Figure imgb0015
    (where M equals 10.)
  • Table 6 provides the totals from Table 5. For all values of Pt, except Pt = 0.9, the original sentence is selected over the candidate. When Pt = 0.9, the candidate is selected over the original.
    Figure imgb0016
    Figure imgb0017
  • EXAMPLE IV
  • In this example, the input word string Wi is "I summit that is what is happening in this case." The first word W1 whose spelling is being checked is "summit". The word "summit" has two simple misspellings: "submit" or "summit".
  • The second word W2 is selected to be "submit". Therefore, the candidate word string Wc is "I submit that is what is happening in this case."
  • Table 7 shows the logarithms of the estimated probabilities of the trigrams and of correctly spelling or incorrectly spelling each word. Since M=2, the probability P(W1|W2)= P t 2
    Figure imgb0018
    .
  • Table 8 provides the totals from Table 7. For all values of Pt, the candidate sentence is selected over the original typed sentence. A correction is therefore made in all cases.
    Figure imgb0019
    Figure imgb0020

Claims (26)

  1. A computerized method of detecting an error in an information signal, said method comprising the steps of:
    providing an input string of information signals (Wi = W1WMWN), each information signal representing information;
    automatically changing a first information signal (W1) in the input string to form a second information signal (W2) representing information different from the information represented by the first information signal (W1), to form a candidate string of information signals (Wc=W2WMWN);
    automatically estimating the probability P(Wi) of occurrence of the input string (Wi = W1WMWN) of information signals on the basis of a body of information signals;
    automatically estimating the probability P(Wc) of occurrence of the candidate string (Wc = W2WMWN) of information signals on the basis of the body of information signals;
    automatically estimating the probability P(Wi|Wc) of misrepresenting the information represented by the candidate string (Wc = W2WMWN) of information signals as the input string (Wi = W1WMWN) of information signals;
    comparing P(Wi) with the product P(Wc)P(Wi|Wc); and outputting a first error signals if P(Wi) is greater than P(Wc)P(Wi|Wc), or outputting a second error signal if P(Wi) is less than P(Wc)P(Wi|Wc).
  2. A method as claimed in Claim 1, characterized in that:
    the first output signal comprises the input string of information signals;
    the second output signal comprises the candidate string of information signals; and
    the probability P(Wi|Wc) is estimated as the probability P(W1|W2) of misrepresenting the information represented by the second information signal W2 as the first information signal W1.
  3. A method as claimed in Claim 2, characterized in that:
    the method further comprises the step of providing a set of words, each word having a spelling;
    each information signal in the input string of information signals represents a word which is a member of the set of words; and
    the second information signal W2 represents a word which is a member of the set of words, the word represented by the second information signal being different from the word being represented by the first information signal.
  4. A method as claimed in Claim 3, characterized in that:
    the method further comprises the step of estimating the probability P(Wi|Wi) of correctly representing the information represented by all of the information signals in the input string of information signals Wi ;
    the step of comparing comprises comparing the product P(Wi)P(Wi|Wi) with the product P(Wc)P(Wi|Wc); and
    the step of outputting comprises outputting the first output signal if P(Wi)P(Wi|Wi) is greater than P(Wc)P(Wi|Wc), or outputting the second output signal if P(Wi)P(Wi|Wi) is less than P(Wc)P(Wi|Wc).
  5. A method as claimed in Claim 4, characterized in that: the probability P(Wi|Wi) is estimated as the probability P(W1|W1) of correctly representing the information represented by the first information signal W1.
  6. A method as claimed in Claim 5, characterized in that the step of changing the first information signal W1 to form the second information signal W2 comprises:
    adding a letter to the word represented by the first information signal to form a tentative word;
    comparing the tentative word to each word in the set of words; and
    representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  7. A method as claimed in Claim 5, characterized in that the step of changing the first information signal W1 to form the second information signal W2 comprises:
    deleting a letter from the word represented by the first information signal to form a tentative word;
    comparing the tentative word to each word in the set of words; and
    representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  8. A method as claimed in Claim 5, characterized in that:
    the first information signal represents a word having at least two letters; and
    the step of changing the first information signal W1 to form the second information signal W2 comprises:
    transposing at least two letters in the word represented by the first information signal to form a tentative word;
    comparing the tentative word to each word in the set of words; and
    representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  9. A method as claimed in Claim 5, characterized in that:
    the first information signal represents a word having at least one letter; and
    the step of changing the first information signal W1 to form the second information signal W2 comprises:
    replacing a letter in the word represented by the first information signal to form a tentative word;
    comparing the tentative word to each word in the set of words; and
    representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  10. A method as claimed in Claim 5, characterized in that the step of changing the first information signal W1 to form the second information signal W2 comprises:
    identifying a confusion group of M different words in the set of words, each word in the confusion group having a spelling which differs from the spelling of the word represented by the first information signal by no more than two letters; and
    representing one word in the confusion group as the second information signal W2.
  11. A method as claimed in Claim 5, characterized in that the step of changing the first information signal W1 to form the second information signal W2 comprises:
    identifying a confusion group of M different words in the set of words, each word in the confusion group being confusable with the word represented by the first information signal; and
    representing one word in the confusion group as the second information signal W2.
  12. A method as claimed in Claim 11, characterized in that:
    the probability P(W1|W1) is estimated to be 0.999; and
    the probability P(Wi|Wc) is estimated to be 0.001 M
    Figure imgb0021
    .
  13. A method as claimed in Claim 1, characterized in that the second output signal comprises an error indication signal.
  14. A computerized apparatus for detecting an error in an information signal, said apparatus comprising:
    means for providing an input string of information signals (Wi = W1WMWN), each information signal representing information;
    means for automatically changing a first information signal (W1) in the input string to form a second information signal (W2) representing information different from the information represented by the first information signal, to form a candidate string of information signals (Wc=W2WMWN);
    means for automatically estimating the probability P(Wi) of occurrence of the input string (Wi = W1WMWN) of information signals on the basis of a body of information signals;
    means for automatically estimating the probability P(Wc) of occurrence of the candidate string (Wc=W2WMWN) of information signals on the basis of the body of information signals;
    means for automatically estimating the probability P(Wi|Wc) of misrepresenting the information represented by the candidate string (Wc=W2WMWN) of information signals as the input string (Wi = W1WMWN) of information signals;
    means for comparing P(Wi) with the product P(Wc)P(Wi|Wc); and
    means for outputting a first error signals if P(Wi) is greater than P(Wc)P(Wi|Wc), or outputting a second error signal if P(Wi) is less than P(Wc)P(Wi|Wc).
  15. An apparatus as claimed in Claim 14, characterized in that:
    the first output signal comprises the input string of information signals;
    the second output signal comprises the candidate string of information signals; and
    the probability P(Wi|Wc) is estimated as the probability P(W1|W2) of misrepresenting the information represented by the second information signal W2 as the first information signal W1.
  16. An apparatus as claimed in Claim 15, characterized in that:
    the apparatus further comprises dictionary means for storing a set of words, each word having a spelling;
    each information signal in the input string of information signals represents a word which is a member of the set of words; and
    the second information signal W2 represents a word which is a member of the set of words, the word represented by the second information signal being different from the word being represented by the first information signal.
  17. An apparatus as claimed in Claim 16, characterized in that:
    the apparatus further comprises means for estimating the probability P(Wi|Wi) of correctly representing all of the information represented by the input string of information signals Wi ;
    the means for comparing comprises means for comparing the product P(Wi)P(Wi|Wi) with the product P(Wc)P(Wi|Wc) ; and
    the means for outputting comprises means for outputting the first output signal if P(Wi)P(Wi|Wi) is greater than P(Wc)P(Wi|Wc), or outputting the second output signal if P(Wi)P(Wi|Wi) is less than P(Wc)P(Wi|Wc).
  18. An apparatus as claimed in Claim 17, characterized in that:
    the probability P(Wi|Wi) is estimated as the probability P(W1|W1) of correctly representing the information represented by the first information signal W1.
  19. An apparatus as claimed in Claim 18, characterized in that the means for changing the first information signal W1 to form the second information signal W2 comprises:
    means for adding a letter to the word represented by the first information signal to form a tentative word;
    means for comparing the tentative word to each word in the set of words; and
    means for representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  20. An apparatus as claimed in Claim 18, characterized in that the means for changing the first information signal W1 to form the second information signal W2 comprises:
    means for deleting a letter from the word represented by the first information signal to form a tentative word;
    means for comparing the tentative word to each word in the set of words; and
    means for representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  21. An apparatus as claimed in Claim 18, characterized in that
    the first information signal represents a word having at least two letters; and
    the means for changing the first information signal W1 to form the second information signal W2 comprises:
    means for transposing at least two letters in the word represented by the first information signal to form a tentative word;
    means for comparing the tentative word to each word in the set of words; and
    means for representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  22. An apparatus as claimed in Claim 18, characterized in that:
    the first information signal represents a word having at least one letter; and
    the means for changing the first information signal W1 to form the second information signal W2 comprises:
    means for replacing a letter in the word represented by the first information signal to form a tentative word;
    means for comparing the tentative word to each word in the set of words; and
    means for representing the tentative word as the second information signal W2 if the tentative word matches a word in the set of words.
  23. An apparatus as claimed in Claim 18, characterized in that the means for changing the first information signal W1 to form the second information signal W2 comprises:
    means for identifying a confusion group of M different words in the set of words, each word in the confusion group having a spelling which differs from the spelling of the word represented by the first information signal by no more than two letters; and
    means for representing one word in the confusion group as the second information signal W2.
  24. An apparatus as claimed in Claim 18, characterized in that the means for changing the first information signal W1 to form the second information signal W2 comprises:
    means for identifying a confusion group of M different words in the set of words, each word in the confusion group being confusable with the word represented by the first information signal; and
    means for representing one word in the confusion group as the second information signal W2.
  25. An apparatus as claimed in Claim 24, characterized in that:
    the probability P(W1|W1) is estimated to be 0.999; and
    the probability P(Wi|Wc) is estimated to be 0.001 M
    Figure imgb0022
    .
  26. An apparatus as claimed in Claim 14, characterized in that the second output signal comprises an error indication signal.
EP19900108508 1989-08-31 1990-05-07 Method and apparatus for spelling error detection and correction Expired - Lifetime EP0415000B1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US07/401,584 US5258909A (en) 1989-08-31 1989-08-31 Method and apparatus for "wrong word" spelling error detection and correction
US401584 1989-08-31

Publications (3)

Publication Number Publication Date
EP0415000A2 EP0415000A2 (en) 1991-03-06
EP0415000A3 EP0415000A3 (en) 1992-01-02
EP0415000B1 true EP0415000B1 (en) 1997-07-23

Family

ID=23588329

Family Applications (1)

Application Number Title Priority Date Filing Date
EP19900108508 Expired - Lifetime EP0415000B1 (en) 1989-08-31 1990-05-07 Method and apparatus for spelling error detection and correction

Country Status (4)

Country Link
US (1) US5258909A (en)
EP (1) EP0415000B1 (en)
JP (1) JPH079655B2 (en)
DE (1) DE69031099D1 (en)

Families Citing this family (60)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5604897A (en) * 1990-05-18 1997-02-18 Microsoft Corporation Method and system for correcting the spelling of misspelled words
US5572423A (en) * 1990-06-14 1996-11-05 Lucent Technologies Inc. Method for correcting spelling using error frequencies
US5157759A (en) * 1990-06-28 1992-10-20 At&T Bell Laboratories Written language parser system
KR950008022B1 (en) * 1991-06-19 1995-07-24 가나이 쯔또무 Charactor processing method and apparatus therefor
DE4323241A1 (en) * 1993-07-12 1995-02-02 Ibm Method and computer system to search faulty strings in a text
US5537317A (en) * 1994-06-01 1996-07-16 Mitsubishi Electric Research Laboratories Inc. System for correcting grammer based parts on speech probability
US5485372A (en) * 1994-06-01 1996-01-16 Mitsubishi Electric Research Laboratories, Inc. System for underlying spelling recovery
US5761689A (en) * 1994-09-01 1998-06-02 Microsoft Corporation Autocorrecting text typed into a word processing document
JP2809341B2 (en) * 1994-11-18 1998-10-08 松下電器産業株式会社 Information summary method, information summarizing apparatus, weighting method, and teletext receiving apparatus.
JPH08235182A (en) * 1995-02-28 1996-09-13 Canon Inc Method and device for document processing
US5828991A (en) * 1995-06-30 1998-10-27 The Research Foundation Of The State University Of New York Sentence reconstruction using word ambiguity resolution
GB2314433A (en) * 1996-06-22 1997-12-24 Xerox Corp Finding and modifying strings of a regular language in a text
US5956739A (en) * 1996-06-25 1999-09-21 Mitsubishi Electric Information Technology Center America, Inc. System for text correction adaptive to the text being corrected
US5999896A (en) * 1996-06-25 1999-12-07 Microsoft Corporation Method and system for identifying and resolving commonly confused words in a natural language parser
WO1998020428A1 (en) * 1996-11-01 1998-05-14 Bland Linda M Interactive and automatic processing of text to identify language bias
US6064959A (en) * 1997-03-28 2000-05-16 Dragon Systems, Inc. Error correction in speech recognition
US6047300A (en) * 1997-05-15 2000-04-04 Microsoft Corporation System and method for automatically correcting a misspelled word
US6016467A (en) * 1997-05-27 2000-01-18 Digital Equipment Corporation Method and apparatus for program development using a grammar-sensitive editor
US6782510B1 (en) 1998-01-27 2004-08-24 John N. Gross Word checking tool for controlling the language content in documents using dictionaries with modifyable status fields
US6963871B1 (en) * 1998-03-25 2005-11-08 Language Analysis Systems, Inc. System and method for adaptive multi-cultural searching and matching of personal names
US8812300B2 (en) 1998-03-25 2014-08-19 International Business Machines Corporation Identifying related names
US8855998B2 (en) 1998-03-25 2014-10-07 International Business Machines Corporation Parsing culturally diverse names
US6424983B1 (en) 1998-05-26 2002-07-23 Global Information Research And Technologies, Llc Spelling and grammar checking system
US6131102A (en) * 1998-06-15 2000-10-10 Microsoft Corporation Method and system for cost computation of spelling suggestions and automatic replacement
US6175834B1 (en) * 1998-06-24 2001-01-16 Microsoft Corporation Consistency checker for documents containing japanese text
US6401060B1 (en) * 1998-06-25 2002-06-04 Microsoft Corporation Method for typographical detection and replacement in Japanese text
US6618697B1 (en) * 1999-05-14 2003-09-09 Justsystem Corporation Method for rule-based correction of spelling and grammar errors
US7403888B1 (en) * 1999-11-05 2008-07-22 Microsoft Corporation Language input user interface
US7165019B1 (en) 1999-11-05 2007-01-16 Microsoft Corporation Language input architecture for converting one text form to another text form with modeless entry
US6848080B1 (en) 1999-11-05 2005-01-25 Microsoft Corporation Language input architecture for converting one text form to another text form with tolerance to spelling, typographical, and conversion errors
US7047493B1 (en) * 2000-03-31 2006-05-16 Brill Eric D Spell checker with arbitrary length string-to-string transformations to improve noisy channel spelling correction
US7398467B1 (en) * 2000-06-13 2008-07-08 International Business Machines Corporation Method and apparatus for providing spelling analysis
DE60222413T2 (en) * 2001-04-19 2008-06-12 British Telecommunications P.L.C. voice recognition
US7076731B2 (en) * 2001-06-02 2006-07-11 Microsoft Corporation Spelling correction system and method for phrasal strings using dictionary looping
US6560559B2 (en) * 2001-08-17 2003-05-06 Koninklijke Philips Electronics N.V. System and method for detecting and correcting incorrect hand position of a computer user
EP1288790A1 (en) * 2001-08-29 2003-03-05 Tarchon BV Method of analysing a text corpus and information analysis system
US7613601B2 (en) * 2001-12-26 2009-11-03 National Institute Of Information And Communications Technology Method for predicting negative example, system for detecting incorrect wording using negative example prediction
US20040002850A1 (en) * 2002-03-14 2004-01-01 Shaefer Leonard Arthur System and method for formulating reasonable spelling variations of a proper name
US7440941B1 (en) * 2002-09-17 2008-10-21 Yahoo! Inc. Suggesting an alternative to the spelling of a search query
AU2003297295A1 (en) 2002-12-30 2004-07-29 Fannie Mae System and method of processing data pertaining to financial assets
US20040128230A1 (en) 2002-12-30 2004-07-01 Fannie Mae System and method for modifying attribute data pertaining to financial assets in a data processing system
US7885889B2 (en) * 2002-12-30 2011-02-08 Fannie Mae System and method for processing data pertaining to financial assets
US8543378B1 (en) 2003-11-05 2013-09-24 W.W. Grainger, Inc. System and method for discerning a term for an entry having a spelling error
US7672927B1 (en) 2004-02-27 2010-03-02 Yahoo! Inc. Suggesting an alternative to the spelling of a search query
US7254774B2 (en) * 2004-03-16 2007-08-07 Microsoft Corporation Systems and methods for improved spell checking
US7406416B2 (en) * 2004-03-26 2008-07-29 Microsoft Corporation Representation of a deleted interpolation N-gram language model in ARPA standard format
US20070005586A1 (en) * 2004-03-30 2007-01-04 Shaefer Leonard A Jr Parsing culturally diverse names
US7478038B2 (en) * 2004-03-31 2009-01-13 Microsoft Corporation Language model adaptation using semantic supervision
US7634741B2 (en) * 2004-08-31 2009-12-15 Sap Ag Method and apparatus for managing a selection list based on previous entries
US20060293890A1 (en) * 2005-06-28 2006-12-28 Avaya Technology Corp. Speech recognition assisted autocompletion of composite characters
US7664629B2 (en) * 2005-07-19 2010-02-16 Xerox Corporation Second language writing advisor
US8249873B2 (en) * 2005-08-12 2012-08-21 Avaya Inc. Tonal correction of speech
US20070214189A1 (en) * 2006-03-10 2007-09-13 Motorola, Inc. System and method for consistency checking in documents
EP1855210B1 (en) * 2006-05-11 2018-01-03 Dassault Systèmes Spell checking
US7818332B2 (en) * 2006-08-16 2010-10-19 Microsoft Corporation Query speller
US7877375B1 (en) * 2007-03-29 2011-01-25 Oclc Online Computer Library Center, Inc. Name finding system and method
WO2009016631A2 (en) * 2007-08-01 2009-02-05 Ginger Software, Inc. Automatic context sensitive language correction and enhancement using an internet corpus
US8700997B1 (en) * 2012-01-18 2014-04-15 Google Inc. Method and apparatus for spellchecking source code
CN103578464B (en) * 2013-10-18 2017-01-11 威盛电子股份有限公司 To establish a method of language model, speech recognition method and electronic device
KR20160069329A (en) 2014-12-08 2016-06-16 삼성전자주식회사 Method and apparatus for training language model, method and apparatus for recognizing speech

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA1182570A (en) * 1982-04-30 1985-02-12 Frederick R. Lange System for detecting and correcting contextual errors in a text processing system

Also Published As

Publication number Publication date
DE69031099D1 (en) 1997-09-04
EP0415000A2 (en) 1991-03-06
JPH079655B2 (en) 1995-02-01
US5258909A (en) 1993-11-02
JPH0398158A (en) 1991-04-23
EP0415000A3 (en) 1992-01-02

Similar Documents

Publication Publication Date Title
Collins et al. Prepositional phrase attachment through a backed-off model
Sproat et al. A statistical method for finding word boundaries in Chinese text
US7917355B2 (en) Word detection
US7853874B2 (en) Spelling and grammar checking system
Mikheev Automatic rule induction for unknown-word guessing
CN1171199C (en) Information retrieval and speech recognition based on language models
CN1143232C (en) Automatic segmention of text
US7584093B2 (en) Method and system for generating spelling suggestions
Palmer A trainable rule-based algorithm for word segmentation
KR101146539B1 (en) Systems and methods for spell correction of non-roman characters and words
US7680649B2 (en) System, method, program product, and networking use for recognizing words and their parts of speech in one or more natural languages
US8386240B2 (en) Domain dictionary creation by detection of new topic words using divergence value comparison
US5956739A (en) System for text correction adaptive to the text being corrected
CN1954315B (en) Systems and methods for translating chinese pinyin to chinese characters
Denis et al. Coupling an annotated corpus and a morphosyntactic lexicon for state-of-the-art POS tagging with less human effort
US4833610A (en) Morphological/phonetic method for ranking word similarities
JP3095552B2 (en) How to search for documents related to the same topic
US5541836A (en) Word disambiguation apparatus and methods
Whitelaw et al. Using the web for language independent spellchecking and autocorrection
US6694055B2 (en) Proper name identification in chinese
US6848080B1 (en) Language input architecture for converting one text form to another text form with tolerance to spelling, typographical, and conversion errors
US7689409B2 (en) Text language identification
US7165019B1 (en) Language input architecture for converting one text form to another text form with modeless entry
EP0651340B1 (en) Language translation apparatus and method using context-based translation models
US7113950B2 (en) Automated error checking system and method

Legal Events

Date Code Title Description
AK Designated contracting states:

Kind code of ref document: A2

Designated state(s): DE FR GB IT

17P Request for examination filed

Effective date: 19901213

AK Designated contracting states:

Kind code of ref document: A3

Designated state(s): DE FR GB IT

17Q First examination report

Effective date: 19940817

AK Designated contracting states:

Kind code of ref document: B1

Designated state(s): DE FR GB IT

PG25 Lapsed in a contracting state announced via postgrant inform. from nat. office to epo

Ref country code: FR

Effective date: 19970723

Ref country code: IT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 19970723

REF Corresponds to:

Ref document number: 69031099

Country of ref document: DE

Date of ref document: 19970904

PG25 Lapsed in a contracting state announced via postgrant inform. from nat. office to epo

Ref country code: DE

Effective date: 19971024

EN Fr: translation not filed
26N No opposition filed
PGFP Postgrant: annual fees paid to national office

Ref country code: GB

Payment date: 20000427

Year of fee payment: 11

PG25 Lapsed in a contracting state announced via postgrant inform. from nat. office to epo

Ref country code: GB

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20010507

GBPC Gb: european patent ceased through non-payment of renewal fee

Effective date: 20010507